Micro Kernel

Micro-kernels are small, highly optimized code routines, often written in assembly language or using intrinsic functions, that form the core of many high-performance linear algebra operations, particularly matrix multiplication (GEMM). Current research focuses on automating the generation of these micro-kernels, moving away from manual, architecture-specific coding towards compiler-based approaches that leverage frameworks like TVM and Exo. This automation improves portability, maintainability, and allows for easier adaptation to various hardware architectures and data types, ultimately leading to more efficient and widely applicable scientific computing libraries.